Pomelli is a novel AI experiment developed by Google Labs in partnership with Google DeepMind, designed to empower small-to-medium-sized businesses (SMBs) by facilitating the creation of scalable, on-brand social media campaigns. It addresses the significant challenges SMBs face in generating impactful, branded content due to constraints in time, budget, and design expertise.

The core methodology of Pomelli is structured around a three-step AI-driven process:

1. Build your business DNA: This initial phase involves the automatic extraction and understanding of a business's unique brand identity. Upon receiving a user-provided website URL, Pomelli employs advanced analytical techniques. This likely includes:
   • Web Scraping and Content Analysis: To parse text content, identify key messaging, and infer the brand's tone of voice. This would involve Natural Language Processing (NLP) models for sentiment analysis, style detection (e.g., formal, informal, playful), and keyword extraction.
   • Image Analysis and Computer Vision: To process existing images on the website. This would involve algorithms for color palette extraction, font identification from visual elements, and potentially style recognition to understand the visual aesthetic (e.g., minimalist, vibrant, professional).

The aggregate of these extracted features—comprising tone of voice, custom fonts, dominant imagery characteristics, and color palette—forms a structured "Business DNA" profile. This profile serves as a foundational constraint, ensuring all subsequent AI-generated content adheres strictly to the brand's established identity, promoting authenticity and consistency across various marketing channels.

2. Generate tailored campaign ideas: Leveraging the established Business DNA, Pomelli proceeds to generate strategic campaign concepts. This step aims to alleviate the ideation burden for SMBs. The process involves:
   • Generative AI (e.g., Large Language Models - LLMs): The Business DNA (including inferred target audience, product/service descriptions, and brand values) is provided as contextual input to a generative AI model. This model then synthesizes novel, relevant campaign ideas.
   • Prompt Engineering: Users have the option to provide a custom prompt, allowing them to guide the ideation process more precisely, influencing the LLM's output to align with specific marketing objectives or visions. The model then generates diverse campaign focuses, from which the user can quickly select the most suitable.

3. Edit and create high-quality, branded creatives: In the final stage, Pomelli translates the selected campaign idea into a suite of high-quality, on-brand marketing assets suitable for social media, websites, and advertising platforms. This involves multi-modal generative AI capabilities:
   • Text Generation: Based on the campaign idea and Business DNA, an LLM generates various forms of copy, such as social media captions, ad headlines, and body text, ensuring the tone of voice is consistent.
   • Image Generation: Utilizing image generative models (e.g., Diffusion Models), visual assets are created. These models are conditioned on the Business DNA's visual elements, applying the specified color palettes, fonts (potentially through text-to-image synthesis or post-processing), and overall image style to ensure visual brand coherence.

Users retain full control, enabling them to browse and select preferred assets, and make direct edits to both text and images within the tool. All generated assets are provided in a downloadable format, ready for immediate deployment across the business's digital channels.

Pomelli is currently available as a public beta experiment in the United States, Canada, Australia, and New Zealand, providing an early-stage, AI-driven solution to democratize sophisticated content creation for SMBs.